Pan-tumor landscape of fibroblast growth factor receptor 1-4 genomic alterations

Background Selective tyrosine kinase inhibitors targeting fibroblast growth factor receptor (FGFR) 1-4 genomic alterations are in development or have been approved for FGFR-altered cancers (e.g. bladder cancer and advanced intrahepatic cholangiocarcinoma). Understanding FGFR inhibitor-resistance mechanisms is increasingly relevant; we surveyed the pan-tumor landscape of FGFR1-4 genomic alterations [short variants (SVs), gene rearrangements (REs), and copy number alterations (CNAs)], including their association with tumor mutational burden (TMB) and the genomic comutational landscape. Patients and methods Comprehensive genomic profiling of 355 813 solid tumor clinical cases was performed using the FoundationOne and FoundationOne CDx assays (Foundation Medicine, Inc.) to identify genomic alterations in >300 cancer-associated genes and TMB (determined on ≤1.1 megabases of sequenced DNA). Results FGFR1-4 SVs and REs occurred in 9603/355 813 (2.7%), and CNAs in 15 078/355 813 (4.2%) samples. Most common FGFR alterations for bladder cancer, intrahepatic cholangiocarcinoma, and glioma were FGFR3 SVs (1051/7739, 13.6%), FGFR2 REs (618/6641, 9.3%), and FGFR1 SVs (239/11 550, 2.1%), respectively. We found several, potentially clinically relevant, tumor-specific associations between FGFR1-4 genomic alterations and other genomic markers. FGFR3 SV-altered bladder cancers and FGFR1 SV-altered gliomas were significantly less likely to be TMB-high versus unaltered samples. FGFR3 SVs in bladder cancer significantly co-occurred with TERT and CDKN2A/B alterations; TP53 and RB1 alterations were mutually exclusive. In intrahepatic cholangiocarcinoma, FGFR2 REs significantly co-occurred with BAP1 alterations, whereas KRAS, TP53, IDH1, and ARID1A alterations were mutually exclusive. FGFR1 SVs in gliomas significantly co-occurred with H3-3A and PTPN11 alterations, but were mutually exclusive with TERT, EGFR, TP53, and CDKN2A/B alterations. Conclusions Overall, our hypothesis-generating findings may help to stratify patients in clinical trials and guide optimal targeted therapy in those with FGFR alterations.


INTRODUCTION
The fibroblast growth factor receptor (FGFR) family consists of four transmembrane receptor proteins, FGFR1-4. 1 Binding of their respective fibroblast growth factor ligands results in receptor dimerization and activation of downstream signaling pathways [e.g. extracellular-signal regulated kinase (ERK)/ mitogen-activated protein kinase (MAPK)], which promotes cell survival, proliferation, development, angiogenesis, and differentiation. 1 Oncogenic alterations in FGFR1-4, including short variants (SVs), gene rearrangements (REs; i.e. movement of DNA gene sequences across the genome, which may lead to gene fusions), and copy number alterations (CNAs), are found in w7% of all human cancers, most commonly in urothelial, breast, endometrial, squamous lung, and ovarian cancer, and in cholangiocarcinoma. 1,2 The pathogenic potential of FGFR alterations is both alteration-and tumor type-specific, impacting the likelihood of patient response to FGFR inhibitors, and therefore underscoring the need for further delineation to inform clinical decision-making.
Based on positive results from clinical trials, [3][4][5] three selective tyrosine kinase inhibitors targeting particular FGFRs have been granted accelerated approval by the Food and Drug Administration (FDA) for treatment of FGFR-altered cancers, including erdafitinib for FGFR2/3-altered, previously treated, locally advanced, or metastatic urothelial cancer, as well as pemigatinib and infigratinib for previously treated, locally advanced, or metastatic cholangiocarcinoma with FGFR2 REs. [6][7][8] Pemigatinib has also been approved/recommended by the European Medicines Agency (EMA), the National Institute for Health and Care Excellence (NICE), and the Chinese National Medical Products Administration for the same indication, and by the Japanese Ministry of Health, Labour and Welfare for patients with unresectable, FGFR2 RE-altered, biliary tract cancer that has progressed after at least one prior line of systemic therapy. [9][10][11][12] Futibatinib, the pan-FGFR1-4 inhibitor, has been granted accelerated FDA approval for the treatment of advanced cholangiocarcinoma with FGFR2 REs after positive phase II clinical data. 13,14 Novel agents are being tested in clinical trials; for example, the highly selective FGFR2 inhibitor RLY-4008 in a clinical trial of patients with intrahepatic cholangiocarcinoma (and other advanced solid tumors) and FGFR2 alterations (NCT04526106). Initial results suggest potent and selective FGFR2 inhibition and encouraging antitumor activity in FGFR inhibitor-naïve patients, across all doses. 15 The activity profile of FGFR inhibitors varies according to the type of FGFR alteration, with greater objective response rate and progressionfree survival associated with SVs and REs, compared with CNAs. 16,17 Some patients may also develop resistance to FGFR inhibitors due to mutations arising in the 'gatekeeper' residues of FGFR1-4, which are responsible for controlling access of FGFR inhibitors to the receptor. 1 Resistance can further occur via mutations in other regions of the kinase domain of the receptor: for example, N550H in FGFR2, which acts as a molecular brake that restricts the kinase to an uncontrolled, active state. 18 Clinically validated, comprehensive genomic profiling (CGP)based assays are critical to identify patients who may benefit most from FGFR inhibitors (or other targeted therapies). 1,19 CGP is a next-generation sequencing-based method that detects novel and known variants of the four main classes of genomic alterations (insertions and deletions, REs, CNAs, and substitutions), and genomic signatures such as tumor mutational burden (TMB), microsatellite instability, and genomewide loss of heterozygosity (for patients with ovarian cancer to provide prognostic, diagnostic, and predictive insights that inform research or treatment decisions for individual patients across all cancer types). 20 The incidence of FGFR1-4 genomic alterations across different tumor types and the genomic comutational landscape influencing the response to FGFR inhibitors should be studied in a broad manner.This is particularly valuable for tumor types other than cholangiocarcinoma and urothelial carcinoma, in which FGFR1-4 alterations have been described, but their prevalence and distribution are not well understood. Therefore, we aimed to comprehensively survey the pan-tumor landscape of FGFR1-4 genomic alterations (SVs, REs, and CNAs), including overall and disease-specific prevalence. For commonly occurring FGFR1-4 SVs and REs, we also describe their tumor type-specific association with TMB and the genomic comutational landscape, including significantly co-occurring and mutually exclusive (i.e. significantly likely to not co-occur) genomic alterations. This should help to inform molecular-based patient stratification for future clinical trials, next-generation FGFR inhibitor development, and combination therapy for FGFR-altered tumors.

PATIENTS AND METHODS
CGP of 355 813 solid tumor clinical cases (as diagnosed by the treating physician and confirmed on hematoxylin-and eosinstained slides) was performed using the FoundationOne (F1) and F1CDx assays (Foundation Medicine, Inc., Cambridge, MA, USA), as described previously, 21,22 in a Clinical Laboratory Improvement Amendments-certified and College of American Pathologists-accredited laboratory. Regarding urinary tract cancer, we assessed tumors originating from the bladder, urethra, and the upper urinary tract, including urothelial carcinoma and variant histologies. Whereas bladder cancer refers to all cancers originating from the bladder, urinary tract cancers are those originating from areas of the urinary tract other than the bladder, such as the ureter, urethra, and urachus regions.
All samples submitted for sequencing featured a minimum of 20% tumor cell nuclear area and yielded a minimum of 50 ng of extracted DNA. CGP was performed on hybrid-capture, adapter ligation-based libraries, to identify genomic alterations [base substitutions, small insertions and deletions, CNAs (gene copy number of ! specimen ploidy þ4), and REs] in coding exons (F1CDx: n ¼ 309; F1: n ¼ 395) and select introns of cancerassociated genes (F1CDx: n ¼ 36; F1: n ¼ 31), and TMB. 23 TMB was calculated as the number of nondriver somatic coding mutations per megabase (mut/Mb) of genome sequenced; TMB-high was defined as !10 mut/ Mb and TMB-low as <10 mut/Mb. All genomic alterations studied included only those described as functional or pathogenic in the literature and seen in the Catalogue Of Somatic Mutations In Cancer (COSMIC) repository, 24 or those with a likely functional status (frameshift or truncation events in tumor suppressor genes). Variants of unknown significance were not studied.
As self-reported race was not available, genomic ancestry was determined for each patient sample. For each profiling platform (F1 and F1CDx), >40 000 single-nucleotide polymorphism sites sequenced by CGP were identified. To remove biases due to linkage disequilibrium, linkage pruning was performed using PLINK (using the eindep flag with a window size of 50, a step size of 5, and a variance inflation factor threshold of 2). A random forest classifier was trained on the 1000 Genomes Project samples to identify ancestral populations [African, Admixed American (Hispanic), East Asian, South Asian] using genetic variation at the singlenucleotide polymorphism sites. Genetic variation was defined by 10 features that captured allele-count variation as determined by principal component analysis. This classifier was applied to CGP patient samples to assign them to one of the ancestral populations.
All statistical analyses were performed using R software v4.0.3 (R Foundation for Statistical Computing, Vienna, Austria) and Python v.2.7.16 (Python Software Foundation, Wilmington, DE, USA). Proportions of categorical variables were compared using Fisher's exact test. Wilcoxon rank sum was used to test for differences between continuous variables. All P values were two-sided, and multiple hypothesis testing correction was performed using the BenjaminieHochberg procedure to calculate the false discovery rate.
Approval for this analysis, including a waiver of informed consent and a Health Insurance Portability and Accountability Act waiver of authorization, was obtained from the Western Institutional Review Board (Protocol No. 20152817).

Comutational landscape and mutual exclusivity of FGFR1-4 genomic alterations
We studied the baseline demographics of commonly occurring FGFR SVs and REs in bladder cancer, cholangiocarcinoma, and glioma. These were selected based on the prevalence of FGFR SVs and REs and emerging clinical interest (e.g. for gliomas). We also assessed the diseasespecific alteration type, comutational landscape, and mutual exclusivity of these cancer types, as well as urinary tract cancer. In bladder cancer, genomic alterations were most common in FGFR3, particularly in the form of SVs. FGFR3 SV-altered bladder cancers were significantly more likely to be from older patients (P ¼ 4.5 Â 10 À5 ; Supplementary    Figure 3A); alterations in TP53 (310/1051, 29.5%, versus 4456/6688, 66.6%; odds ratio ¼ 0.21; P ¼ 3.94 Â 10 À114 ) and RB1 (27/1051, 2.6%, versus 1588/6688, 23.7%; odds ratio ¼ 0.08; P ¼ 1.19 Â 10 À76 ) were found to be significantly mutually exclusive with FGFR3 SVs (Figure 3A). FGFR3 SVs were also common in urinary tract cancer and FGFR3 SV-altered urinary tract cancers were significantly more likely to be TMB-high versus unaltered samples [  were similar to those found for FGFR3 SV-altered bladder cancers (Supplementary Figure S4B,  ) were found to be significantly mutually exclusive ( Figure 3B).

DISCUSSION
To our knowledge, this study is the largest to date that comprehensively describes the pan-tumor landscape of FGFR1-4 genomic alterations (SVs, REs, and CNAs), including overall and disease-specific prevalence and disease-specific genomic comutational landscape. In this study, FGFR1-4 SVs/REs were seen in 2.7% of samples (mostly SVs; FGFR4 alterations made up only a small proportion). As reported previously, 2,25,26 CNAs were detected in a higher proportion (4.2%) and were most common in FGFR1, and SVs/REs were most common in FGFR2/3.
Overall, in our study, samples with FGFR1-4 SVs and REs, versus the unaltered samples, were more likely to be of European ancestry. This may represent a general trend, especially given that FGFR2 REs are more common in patients of European versus Asian descent, 19 although further study with detailed clinical data is required to elucidate whether this impacts patient outcomes. Furthermore, there are more significant socioeconomic variables involved, so it is difficult to draw conclusions.
Consistent with previous studies, 2,26 we found that CNAs occurred most commonly in breast cancer. The usefulness of CNAs as a patient selection tool for FGFR inhibitors is still not well understood, particularly considering the low concordance rate between FGFR1 amplification, the most common type of FGFR CNA, and FGFR1 messenger RNA (and thus protein) expression. 27 However, a translational clinical trial demonstrated high-level FGFR2 amplification to predict response to the FGFR inhibitor AZD4547 in gastric cancer 28 ; thus the association of CNAs with response to FGFR inhibitors may be gene-and tumor type-specific. In this trial, only high-level CNAs were studied; clinical analyses of the response to inhibitors in patients with highlevel FGFR CNAs are therefore required. 28 FGFR3 alterations were most common in bladder and urinary tract cancer in this study, with the majority being SVs. A high prevalence of FGFR3 SVs in urothelial cancers, with an enrichment in tumors originating from the renal pelvis and ureter, has been detected previously. 2 The anti-FGFR therapy erdafitinib has shown clinical effectiveness in these tumor types and has subsequently been approved by the FDA. 3,6,29 However, the response rates reported with other pan-FGFR inhibitors ( 25%) were substantially lower compared with that of erdafitinib (40%). [29][30][31][32][33] These differences may be partly related to the differing properties of the inhibitors (e.g. erdafitinib has a relatively long halflife) and molecular characteristics of the tumors from the varying patient populations. Indeed, we also found in these cancers a significant co-occurrence of FGFR3 SVs with CDKN2A/B and TERT, as has been found previously, 33 suggesting that it may be possible to stratify patients according to these alterations, but clinical data are required to confirm this hypothesis. Phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha (PIK3CA) alterations have also been detected in FGFR3-altered urothelial cancers 2 ; however, combined phosphoinositide 3-kinase and FGFR inhibition with alpelisib and infigratinib, respectively, in PIK3CA-altered solid tumors has not shown clear evidence of synergistic activity, and potential toxicity is a concern. 34 In our study, FGFR3 SVs in bladder cancer were significantly associated with a lower TMB. This is consistent with the fact that FGFR3 SVs are typically enriched in luminaltype tumors such as urothelial cancer, which are characterized by a depleted T-cell environment and a reduced predicted response to immune checkpoint inhibitors. 35,36 Nonetheless, FGFR3 SVs do not appear to confer resistance to immune checkpoint inhibitors, with multiple studies demonstrating similar outcomes between patients with FGFR3-altered and FGFR3-unaltered tumors. [36][37][38] This suggests that combined FGFR and immune checkpoint inhibition may be clinically useful in FGFR-altered, metastatic urothelial cancer. Indeed, preclinical analysis has shown FGFR inhibitors to broaden the T-cell repertoire in cancer models through increased CD4þ and CD8þ T-cell infiltration; the same analysis also demonstrated immune checkpoint inhibitors to focus this pre-existing T-cell response through clonal expansion. 39 Clinically, early data from the NORSE (NCT03473743) and FORT-2 (NCT03473756) trials of pan-FGFR inhibitors combined with immune checkpoint inhibitors in patients with first-line metastatic, FGFR-altered urothelial cancer have shown signals of potential synergism with manageable safety profiles. 40,41 As TMB has been demonstrated to be a marker of benefit to immune checkpoint inhibitor therapies, 42,43 it is possible that future strategies of combination therapy would demonstrate enriched responses in selected populations of patients with TMB-high tumors. Moreover, clinical studies of patients with metastatic urothelial cancer treated with the pan-FGFR inhibitor erdafitinib have shown higher response rates in those previously treated with cancer immunotherapy, 3,44 suggesting that FGFR inhibitors may be useful in patients with urothelial cancers who do not respond to cancer immunotherapy. Our co-occurring mutational analysis of intrahepatic cholangiocarcinoma revealed FGFR2 REs to associate significantly with BAP1 alterations, as seen previously. 19,45 BAP1 is a tumor suppressor gene that has epigenetic functions in cancer 46 ; however, co-occurring alterations in BAP1 are not prognostic for overall survival, objective response rate, or progression-free survival in patients with cholangiocarcinoma treated with pemigatinib, the selective FGFR1/2/3 inhibitor. 19, 45 We also observed TP53 alterations to be significantly mutually exclusive with FGFR2 REs. In contrast to BAP1, the presence of co-occurring TP53 alterations in tumors has previously been associated with a shorter overall survival and median progression-free survival in patients with cholangiocarcinoma treated with pemigatinib, compared with those without. 19,45 This suggests that patients with alterations in tumor suppressor genes such as TP53 may have worse outcomes with FGFR inhibitors, warranting further research into combination therapy in these patients. This may also apply to bladder cancers and gliomas, in which TP53 alterations were also found in this study to be mutually exclusive with FGFR3 SVs and FGFR1 SVs, respectively.
We found that the most common FGFR alterations in gliomas were FGFR1 SVs, as seen in other studies (along with FGFR3 alterations). 2,25 A recent clinical trial of infigratinib in patients with recurrent/progressive glioma and any FGFR alteration revealed durable disease control lasting >1 year in patients with FGFR1/3 point mutations or FGFR3-TACC3 fusions. 47 This indicates that refined future trials of infigratinib, either alone or in combination with other targeted therapies, may be of interest in patients with gliomas harboring specific FGFR aberrations. In this study, FGFR1 SVs in glioma were found to be co-occurring with H3-3A and mutually exclusive with CDKN2A/2B. The genomic co-occurrence findings here and in other studies therefore suggest that possible combination partners for infigratinib may be those with clinical activity against H3-altered gliomas (e.g. ONC2021). 47,48 It is also interesting that both intrahepatic cholangiocarcinoma and glioma FGFR alterations were significantly associated with alterations in epigenetic modifiers, suggesting a potential mechanistic role in these tumors. This is the largest study thus far to investigate pan-cancer FGFR alterations, with unprecedented coverage of analyzed cases and diverse cohort characteristics, enabling crosscomparison of genomic findings of different tumor types using a single, clinically validated assay. A limitation of the study is the lack of treatment outcome data, preventing analyses of individual patient outcomes and response following therapy. A second limitation is the lack of correlation analyses of circulating tumor DNA with tissue biomarker findings, particularly given that next-generation sequencing-based circulating tumor DNA analysis may be able to reveal additional FGFR alterations and therapeutic targets not shown by some forms of tissue-based testing. 49,50 Other limitations include a lack of a central pathology review of samples, large time frame for sample collection, potential bias toward European ancestry, and lack of association with other important analyses; for example, FGFR3 messenger RNA expression or gene expression findings and molecular subtypes.

Conclusions
In conclusion, we comprehensively surveyed the pan-tumor landscape of FGFR1-4 genomic alterations in this study and identified a number of co-occurring and mutually exclusive alterations in other genes, as well as associations with the genomic signature TMB. Such hypothesis-generating findings may help to stratify patients in clinical trials and guide optimal targeted therapy in those with FGFR alterations. Roche Ltd. HN is an employee of Foundation Medicine, Inc. and has stocks/shares in F. Hoffmann-La Roche Ltd. ARP is a consultant for Foundation Medicine, Inc. SR has received funding for an advisory board/research grant from F. Hoffmann-La Roche Ltd, AbbVie, Merck, Bayer, Incyte, and QED Therapeutics; and has received honoraria from Integrated DNA Technologies. SS is an employee of and has stocks/shares in F. Hoffmann-La Roche Ltd. IMS was an employee of Incyte Corporation. AV has performed a speaker, consultancy, and advisory role for AstraZeneca, Bayer, BMS, BTG, Daiichi Sankyo, Eisai, Eli Lilly and Company, F. Hoffmann-La Roche Ltd, GSK, Imaging Equipment (AAA), Incyte, Ipsen, Merck, MSD, Pierre Fabre, Sanofi, Servier, Sirtex, and Terumo.

DATA SHARING
Consented data that can be released are included in the article and its supplementary files. Patients were not consented for the release of underlying genomic sequence data. Academic researchers can gain access to Foundation Medicine data in this study by contacting the corresponding author and filling out a study review committee form. You and your institution will be required to execute a data transfer agreement. For further questions please reach out to Foundation Medicine, Cambridge, MA's compliance department (compliance@foundationmedicine.com).